Cutting Agent Harness Costs by Up to 3x with Lean Prompts and Tool Schemas
Optimizing the system prompt and tool schema sizes in an agent harness can reduce execution costs by up to 3x without degrading benchmark accuracy when using th
#SWE-bench
TAU-HOME.COM stories tagged SWE-bench.
Graft: Open-Source Context Graph Solves Coding Agent Amnesia, Lifts SWE-bench by 12%p
Graft (@nanonets/graft) builds markdown context graphs without vector DBs, lifting Claude Sonnet 5's SWE-bench solve rate from 54% to 66% while cutting token co